func.func @batchmatmulv2_basic(%arg0: tensor<1x4x2xf32>, %arg1: tensor<3x2x4xf32>) -> tensor<3x4x4xf32> {
// CHECK-LABEL:   func @batchmatmulv2_basic
// CHECK-SAME:        ([[LHS:%.*]]: tensor<1x4x2xf32>, [[RHS:%.*]]: tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
// CHECK:           [[LHSSHAPE:%.*]] = shape.shape_of [[LHS]] : tensor<1x4x2xf32>
// CHECK:           [[RHSSHAPE:%.*]] = shape.shape_of [[RHS]] : tensor<3x2x4xf32>

  %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {T = f32, adj_x = false, adj_y = false, grad_x = false, grad_y = false, device = ""} : (tensor<1x4x2xf32>, tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
  func.return %0 : tensor<3x4x4xf32>
}

  func.return %2 : tensor<4x4x4xf32>
}

  %1 = "tf.AddV2"(%0, %0) : (tensor<1x4x4x8xf32>, tensor<1x4x4x8xf32>) -> tensor<1x4x4x8xf32>
  %2 = "tf.Const"() {value = dense<[0, 3, 1, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %3 = "tf.Transpose"(%1, %2) : (tensor<1x4x4x8xf32>, tensor<4xi32>) -> tensor<1x8x4x4xf32>

  func.return %3 : tensor<1x8x4x4xf32>
}

// CHECK-LABEL: move_transpose_handle_broadcast

// CHECK: %[[TRANSPOSE_2:.+]] = stablehlo.transpose %[[CONV]], dims = [0, 3, 1, 2] : (tensor<1x4x4x8xf32>) -> tensor<1x8x4x4xf32>

// -----

// Tests that the conversion doesn't happen when the input dimension numbers
// are not [b, f, 0, 1].

  // CHECK: %[[RELU:[0-9]*]] = "tf.Relu"(%[[TANH]]) {{.*}} tensor<1x4x4x8xf32>
  // CHECK: %[[RES_TRANSPOSE:[0-9]*]] = "tf.Transpose"(%[[RELU]], %[[RES_PERM]])
  // CHECK: return %[[RES_TRANSPOSE]]

  %0 = "tf.Const"() {value = dense<[0, 3, 1, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %1 = "tf.Transpose"(%arg0, %0) : (tensor<1x4x4x8xf32>, tensor<4xi32>) -> tensor<1x8x4x4xf32>

// CHECK: %[[CUSTOM_AGGREGATOR_1:.+]], {{.*}}, {{.*}}, {{.*}} = "tf.CustomAggregator"(%[[XLA_CALL_MODULE]]) {{.*}} : (tensor<1x4x4x8xf32>) -> (tensor<1x4x4x8xf32>, tensor<f32>, tensor<f32>, tensor<0xi64>)

// CHECK: %[[TRANSPOSE_2:.+]] = stablehlo.transpose %[[CUSTOM_AGGREGATOR_1]], dims = [0, 3, 1, 2] : (tensor<1x4x4x8xf32>) -> tensor<1x8x4x4xf32>

// RUN: tf-opt %s -split-input-file -verify-diagnostics

func.func @out_of_bounds_check(%arg0: tensor<1x4x4x8xf32>) -> tensor<1x4x4x8xf32> {
  %0 = "tf.Const"() {value = dense<[0, 3, 1, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %1 = "tf.Const"() {value = dense<[0, 0x4141, 3, 1]> : tensor<4xi32>} : () -> tensor<4xi32>
  %2 = "tf.Transpose"(%arg0, %0) : (tensor<1x4x4x8xf32>, tensor<4xi32>) -> tensor<1x8x4x4xf32>

static const char *const module_with_add =
    R"(module {
func.func @main(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x4x5xf32>) -> tensor<3x4x5xf32> {
  %0 = "tf.AddV2"(%arg0, %arg1) : (tensor<3x4x5xf32>, tensor<3x4x5xf32>) -> tensor<3x4x5xf32>
  func.return %0 : tensor<3x4x5xf32>
}
}
)";

// Test pass filter.
TEST(DataDumperLoggerConfig, TestPassFilter) {
  mlir::DialectRegistry mlir_registry;

  // CHECK-LABEL: test_batch_matmul_broadcast_to_einsum
  // CHECK: "tf.Einsum"(%arg0, %arg1) <{equation = "...mk,...kn->...mn"}> : (tensor<2x2x4xf32>, tensor<2x4x2xf32>) -> tensor<2x2x2xf32>
  %0 = "tf.BatchMatMul"(%arg0, %arg1) {adj_x = false, adj_y = false} : (tensor<2x2x4xf32>, tensor<2x4x2xf32>) -> tensor<2x2x2xf32>
  func.return %0: tensor<2x2x2xf32>
}